[diegosouzapw]

Summary

Port of decolua/9router#1310 by @noestelar, adapted JS→TS to OmniRoute's StreamCtx-based cursor pipeline.

The Cursor composer-* family encodes its visible reply inside the thinking protobuf field, separated from the (private) chain-of-thought by a final </think> sentinel. OmniRoute was forwarding the whole thinking payload as reasoning_content only, so OpenAI-compatible clients saw an empty message.content and never displayed the reply.

Changes

• New pure helpers in open-sse/executors/cursor.ts:
  • isComposerModel(model) — case-insensitive ^composer(?:-|$)/i after stripping any vendor prefix (cu/composer-2.5 ✓).
  • visibleComposerContentFromThinking(text) — slice after last </think>, trimStart.
  • composerReasoningRemainder(text) — slice before last </think> (the still-private chain-of-thought).
• StreamCtx gets a composerVisibleEmittedLength cursor so streaming only emits the incremental tail per thinking frame.
• processFrame thinking branch: for Composer models, emit the post-</think> suffix as content deltas (mirrored into ctx.totalText so the non-streaming aggregator picks it up); for non-Composer models, the existing reasoning_content contract is preserved.
• buildResponseFromCtx: strips the visible suffix from reasoning_content for Composer, so the visible reply is not duplicated between content and reasoning_content.

Test plan

• New tests/unit/cursor-composer-thinking.test.ts (6 tests, all green):
  • pure-helper contracts (composer detection, suffix/remainder slicing);
  • streaming Composer: incremental visible suffix lands as content deltas, hidden reasoning never leaks as content nor as reasoning_content;
  • non-streaming Composer: ctx.totalText populated from </think>-marked thinking;
  • non-Composer regression: thinking still emits as reasoning_content, totalText stays empty.
• Existing tests/unit/cursor-streaming.test.ts — 20/20 green (no regression).
• npx eslint open-sse/executors/cursor.ts tests/unit/cursor-composer-thinking.test.ts — clean.
• npx tsc -p tsconfig.json --noEmit — no new errors in changed files (5 pre-existing drift errors in unrelated tests).

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.